Research map / 3D vision / Human pose estimation
Multi-level Motion Attention for Human Motion Prediction
Wei Mao, Miaomiao Liu, Mathieu Salzmann, Hongdong Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add post-fusion of joint, part and pose attention predictors (ablation) | Motion attention model · H3.6M long-term average 1000ms | MPJPE (mm): 112.1 -> 110.1 | Improved |
| add post-fusion (ablation) | Motion attention model · H3.6M short-term average 400ms | MPJPE (mm): 58.3 -> 57.3 | Improved |
| add post-fusion (ablation) | Motion attention model · H3.6M Walking 1000ms | MPJPE (mm): 58.1 -> 57.1 | Improved |
| add post-fusion (ablation) | Motion attention model · AMASS-BMLrub 1000ms | MPJPE (mm): 67.2 -> 65.8 | Improved |
| add post-fusion (ablation) | Motion attention model · 3DPW 1000ms | MPJPE (mm): 73.7 -> 71.8 | Improved |
| add motion attention to the LTD predictor (10-frame vs 50-frame observation baseline) | GCN predictor · H3.6M Walking 1000ms | MPJPE (mm): 70.7 -> 57.1 | Improved |
| add motion attention to the GCN predictor | LTD-10-10 · H3.6M long-term average 1000ms | MPJPE (mm): 114.0 -> 110.1 | Improved |
| use 100 observed frames at test time instead of 50 (ablation) | Motion attention model (trained with 50 frames) · H3.6M selected sequences with long-range repetition, 1000ms | MPJPE (mm): 121.1 -> 116.9 | Improved |
| use 100 observed frames at test time instead of 50 (ablation) | Motion attention model · H3.6M selected sequences, 80ms | MPJPE (mm): 10.7 -> 10.7 | No clear effect |
| attention only vs LPJ attention | Pose motion attention · H3.6M 4 actions average 400ms | MPJPE (mm): 41.2 -> 37.9 | Improved |
| replace LTD-10-10 by fused attention model | Motion Att. + Post-fusion · AMASS-BMLrub 1000ms | MPJPE (mm): 91.2 -> 65.8 | Improved |
| fusion vs LTD-10-10 at very short term | Motion Att. + Post-fusion · AMASS-BMLrub 80ms | MPJPE (mm): 10.3 -> 11.0 | Hurt |
Closest papers by meaning
- History Repeats Itself: Human Motion Prediction via Motion Attention (ECCV 2020 · similarity 0.97)
- Context-Aware Human Motion Prediction (CVPR 2020 · similarity 0.92)
- Progressively Generating Better Initial Guesses Towards Next Stages for High-Quality Human Motion Prediction (CVPR 2022 · similarity 0.91)
- Learning Trajectory Dependencies for Human Motion Prediction (ICCV 2019 · similarity 0.91)
- Investigating Pose Representations and Motion Contexts Modeling for 3D Motion Prediction (TPAMI 2022 · similarity 0.91)
- Diverse Human Motion Prediction Guided by Multi-level Spatial-Temporal Anchors (ECCV 2022 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
